LSTM Forecasting of UCL Central House's Office Temperature Using Real-Time BMS Data Under Future Climate Conditions

Abstract

This research tackles the challenge of evaluating and forecasting thermal resilience in ageing non-domestic buildings under climate change, where physics-based models are often impractical. It applies multivariate Long Short-Term Memory (LSTM) neural networks, trained on Building Management System (BMS) data from 2023–2025, to assess UCL Central House. Despite a major retrofit in 2010, recent summer peaks up to 31 °C reveal overheating risks and limitations in current setpoint policies. The proposed multivariate LSTM model, using inputs such as indoor temperature, HVAC operation, occupancy, and weather, achieved high predictive accuracy (R² = 0.88, MAPE = 1.14%). Projections for 2030 indicate frequent indoor temperature exceedances above 27–30 °C, with heatwaves reaching 34 °C and cooling demand rising by up to 60%. Although adaptive occupant behaviour offers some resilience, dependence on mechanical cooling poses risks to energy use, carbon emissions, and well-being. The study highlights the potential of data-driven methods to support retrofit strategies, operational planning, and climate resilience.

More information

Identifiers

ORCID for Stephanie Gauthier: orcid.org/0000-0002-1720-1736

Catalogue record

Export record

Altmetrics